CN108256456A - A kind of finger vein identification method based on multiple features Threshold Fusion - Google Patents
A kind of finger vein identification method based on multiple features Threshold Fusion Download PDFInfo
- Publication number
- CN108256456A CN108256456A CN201810015658.8A CN201810015658A CN108256456A CN 108256456 A CN108256456 A CN 108256456A CN 201810015658 A CN201810015658 A CN 201810015658A CN 108256456 A CN108256456 A CN 108256456A
- Authority
- CN
- China
- Prior art keywords
- curvature
- feature
- formula
- threshold
- fusion
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/14—Vascular patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Collating Specific Patterns (AREA)
- Image Analysis (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a kind of finger vein identification methods based on multiple features Threshold Fusion.The present invention is first with the background area curvature gray feature of the dimensional Gaussian formwork calculation finger venous image curvature extraction image based on multi-standard difference, venosomes curvature gray feature and curvature thin line feature.The matching threshold of background area curvature gray feature and venosomes curvature gray feature is calculated with correlation coefficient process, the matching threshold of curvature thin line feature is calculated with MHD algorithms, a kind of multiple features Threshold Fusion decision algorithm is reintroduced, the threshold value that is individually identified of fusion three of the above feature carries out verification judgement.Algorithm proposed by the present invention based on three kinds of characteristic threshold value fusions, the background area available information included due to combining background area curvature gray feature, unobvious are declined to fuzzy finger venous image recognition performance, therefore sincere substantially less than traditional recognizer based on independent thin line feature, the effective recognizer that can rationally and efficiently utilize are refused under low accuracy of system identification.
Description
Technical field
The invention belongs to living things feature recognition and field of information security technology, are related specifically to finger vein identification technology neck
Domain.
Background technology
Finger vein identification technology belongs to biometrics identification technology, it utilizes finger interior vein topological structure into pedestrian
Body authentication.Become grinding for domestic and international many scholars due to its recognition speed is fast, performance is good and feature is not easy to forge safely
Study carefully hot spot.Finger vein recognition system mainly includes finger vein image acquisition, image preprocessing, feature extraction and characteristic matching
Identification.It is unstable by equipment in gatherer process there is some fingers in view of the technological level of current collecting device
Fixed or acquisition environment influence leads to collected finger venous image there are veinprints to obscure, and venosomes are unstable
The problems such as.Therefore stable available feature is extracted from finger venous image and how algorithm is identified using feature and set
Meter is the key link of finger vein identification technology.
Guan Fengxu et al. is with principal component analysis innovatory algorithm extraction feature, and discrimination is higher, but it is directed to vein image
Feature modified hydrothermal process complexity is high, and extraction time is long.Wang Ke person of outstanding talent et al. first determines vein direction, then is carried with the template of the direction
Take thin line feature.Extraction efficiency is high, and inhibits the noise of other direction template extractions.But its template designed is to low-quality spirogram
As extraction effect is bad.Lin Jian et al. obtains the Hessian matrixes of image, then root using Gaussian template second dervative convolved image
Thin line feature is worth to, and different thicknesses are directed to by changing Gaussian template variance according to the mark and feature of Hessian matrixes
Vein obtains best extraction effect.Gaussian template, to the better adaptability of low-quality image, it is steady to extract result than direction template
It is fixed.The characteristics of Sun Xiaolin is distributed according to vertical vein ridge orientation in image in paddy shape calculates multi-direction upper discrete curvature, according to
Curvature value extracts vein center point.But its curvature estimation uses discrete logarithm, and there are more noises for the feature of extraction.Naoto
Miura et al. finds vein center into line trace according to curvature value, accurate to extract vein center point, but its track algorithm is complicated
Degree is higher.It is better than second dervative to image enhancement effects since curvature combines first derivative and second dervative information.Current
It is most of to be identified using single features in finger vena recognizer, and from vein topological structure carry out algorithm for design when only
Extraction thin line feature or endpoint and intersection feature are identified, and the image that finger vein grain obscures is in feature extraction ring
If section only extracts thin line feature, the information that feature includes can be made few and unstable, when database medium sized vein lines blurred picture
Similarity raising between inhomogeneity finger venous image can be caused by being accumulated to certain amount, identify blurred image user difficult
Or occur accidentally knowing.
Invention content
For algorithm of the existing algorithm such as based on trend pass filtering extraction thin line feature and based on the extraction of Hessian matrixes carefully
Caused by the algorithm of line feature is unstable to blurred picture extraction feature the problem of sincere low, this hair are refused in low accuracy of system identification
It is bright that a kind of finger vena recognizer based on multiple features Threshold Fusion is provided.
The technical solution adopted in the present invention is：
A kind of finger vena recognizer based on multiple features Threshold Fusion, includes the following steps：
The dimensional Gaussian template of S1, the multiple standard deviations of construction, Gaussian template expression formula are as follows：
X and y values are codetermined by two-dimensional Gaussian function standard deviation sigma and Gaussian template Plays difference factor t in formula (1),
A length of p=2 × the t of dimensional Gaussian template square edge of window × σ+1.Therefore x and y values are [- p, p].σ values are variable, and
And decide the size of window, different windows size corresponds to the vein of extraction different thicknesses.
S2, the first-order partial derivative h ' that dimensional Gaussian template is first obtainedx(x,y)h′y(x, y), second-order partial differential coefficient h "xx(x,y)
h″yy(x, y), mixed partial derivative h "xy(x, y), and convolved image matrix obtains image array respectively with this 5 partial derivative templates
5 partial derivative H 'x,H′y,H′xx,H′yy,H′xy, then with directional derivative formula (2), (3) ask the single order of four angles, second order side
To derivative.Reference axis x-axis direction is 0 °, and y-axis is 90 °.
In formula (2) (3)Take 0 °, 45 °, 90 °, 135 °.Directional derivative is substituted into curvature formulations (4), obtains vein image
The corresponding curvature value in all directions.And four direction maximum curvature value is asked to save as the final curvature value of vein image, such as formula
(5):
K is the curvature feature finally extracted in formula (5), is maximum curvature value on four direction.
S3, change standard deviation sigma, the extraction result for choosing suitable multiple standard deviations carries out curvature Fusion Features.Test table
It is best that bright two standard deviations extraction results of selection carry out syncretizing effects.Shown in fusion formula such as formula (6)：
K in formula (6)σ1σ is taken for standard deviation1Dimensional Gaussian template extraction curvature as a result, kσ2σ is taken for standard deviation2Two
The curvature of Gaussian template extraction is tieed up as a result, K is the fusion results that two standard deviations extract result.
S4, curvature gray feature decompose, and normalize to obtain respectively according to positive negative region by curvature feature according to formula (7) (8)
Background curvature gray feature K0 and vein curvature gray feature K1.
In formula (7) (8), K is curvature gray level image, K_min and K_max be respectively by curvature gray level image matrix value according to
Two vectors to sort from small to large and from big to small.The average value for taking preceding 10 values of K_min is normalized as negative region
The upper limit takes the average value of preceding 10 values of K_max to mitigate the exposure mutation pair of image border as the normalized upper limit of positive region
The influence of curvature result.Background curvature gray feature and vein curvature gray feature are that two kinds that the present invention extracts are known for merging
Another characteristic.
S5, the extraction of curvature thin line feature, curvature thin line feature is the feature most simplified in vein image medium sized vein region.It is bent
Rate gray level image K passes through binaryzation, filtering, and refinement obtains thin line feature.This paper binary conversion treatments are using threshold point
Algorithm is cut, threshold value is obtained using the mean value in window and variance, binaryzation is carried out to curvature gray-scale map K, is filtered using connected region
Wave denoising, then obtain curvature thin line feature with refinement method of tabling look-up.
Threshold calculations are individually identified in S6, curvature gray feature
Background curvature gray feature and vein curvature gray feature belong to curvature gray feature, the phase of this paper calculating matrix
Relationship number represents the similarity of curvature gray feature.Related coefficient calculation formula such as formula (9):
K in formula (9)a,KbIt is meant for two by curvature gray feature square Cov and calculates two curvature gray feature Ka,KbAssociation
Variance, Var, which is meant, calculates two curvature gray feature Ka,KbVariance.It is T to remember background curvature gray feature recognition threshold0, it is quiet
Arteries and veins curvature gray feature recognition threshold is T1。
Threshold calculations are individually identified in S7, curvature thin line feature
Curvature thin line feature threshold value using modified Hausdorff distance (modified Hausdorff distance,
MHD) algorithm calculates.For A and two curvature thin line features of B, MHD calculation formula such as formula (10), (11):
T2(A, B)=max (d (A, B), d (B, A)) (11)
D is the minimum average B configuration distance of A to B in formula (10), T in formula (11)2MHD threshold values for A and B.First calculate certain point in A
AmThe all the points B into BnDistance, wherein 1≤n≤N is total to N number of point.Then minimum value is obtained, traverse all M points in A and asks flat
Mean value.It is come again from B to A again, takes the two higher value as the MHD threshold values between A and B.
S8, three kinds of characteristic threshold value fusion recognition algorithms
Since curvature gray feature and curvature thin line feature are the features of different dimensions, threshold calculations mode is different, because
This three kinds of Threshold Fusions sequence is the Threshold Fusion for first carrying out two region curvature gray features, both obtain best weight α and
New threshold value, then obtain weight beta and final threshold value T, expression formula such as formula (12) with curvature thin line feature Threshold Fusion:
T=(1- β) (α T0+(1-α)T1)+βT2 (12)
T in formula (12)0,T1,T2Respectively background curvature gray feature, vein curvature gray feature and curvature thin line feature
Normalization hundred-mark system threshold value, α be threshold value T0,T1Fusion weight, β be threshold value T2With T0,T1Merge the fusion weight of threshold value.
S9, match cognization.It is higher that the higher expression image similarity of threshold value is defined herein, identification verification expression formula such as formula
(13):
T in formula (13)sFor the recognition threshold being calculated according to image library, different accuracy of system identification is corresponding with different identification
Threshold value, T are the matching threshold that two width finger venous images calculate, and may determine that two width finger venous images are according to formula (13)
It is no to derive from same finger.
The present invention has the beneficial effect that：
The present invention proposes a kind of finger vena recognizer based on multiple features Threshold Fusion, first with multi-standard difference
The venosomes curvature gray feature of dimensional Gaussian template extraction finger venous image, background area curvature gray feature and curvature
Thin line feature and use threshold fusion algorithm determine that best fusion three kinds of features of weight fusion of each characteristic threshold value are identified.
Experiment show threshold fusion algorithm proposed by the present invention combine the features of finger venous image different dimensions included it is available
Information after the available information for particularly having merged background area, refuses sincere to calculate than traditional identification based on independent thin line feature
Method has a distinct increment under low accuracy of system identification, and it is that a kind of effective finger is quiet to illustrate multiple features threshold fusion algorithm proposed by the present invention
Arteries and veins recognizer.
Three kinds of features of the finger venous image that the present invention extracts are featured from different dimensions in finger venous image
Different degrees of available information, and the available information that carried threshold fusion algorithm has merged three kinds of features carries out verification judgement, identification
Performance is apparently higher than traditional algorithm identified based on independent thin line feature.Particularly, for finger vein grain blurred picture,
The available information that the thin line feature of traditional algorithm extraction includes is few and unstable, and recognition performance is substantially reduced under low accuracy of system identification,
And the algorithm proposed by the present invention based on the fusion of three kinds of characteristic threshold values, it includes due to combining background area curvature gray feature
Background area available information declines unobvious, therefore refused under low accuracy of system identification to fuzzy finger venous image recognition performance
It is sincere to be substantially less than traditional recognizer based on independent thin line feature.Therefore, the finger based on multiple features Threshold Fusion is quiet
Arteries and veins recognizer is the effective recognizer that a kind of finger vein image information is reasonable and efficiently utilizes.
Description of the drawings
Fig. 1 is a width finger vena original-gray image；
Fig. 2 is the curvature gray level image that the template extraction in 0 ° of direction goes out；
Fig. 3 is the curvature gray level image that the template extraction in 45 ° of directions goes out；
Fig. 4 is the curvature gray level image that the template extraction in 90 ° of directions goes out；
Fig. 5 is the curvature gray level image that the template extraction in 135 ° of directions goes out；
Fig. 6 is the curvature gray level image that the template extraction of standard deviation 1.5 goes out；
Fig. 7 is the curvature gray level image that the template extraction of standard deviation 2.5 goes out；
Fig. 8 is the curvature gray level image blending image of the template extraction of standard deviation 1.5 and 2.5；
Fig. 9 is to merge the venosomes curvature gray feature image that curvature gray level image decomposites；
Figure 10 is to merge the background area curvature gray feature image that curvature gray level image decomposites；
Figure 11 is to merge the curvature thin line feature image that curvature gray level image extracts；
Figure 12 be based on three kinds of features extracting of the present invention be individually identified in algorithm ROC curve and other papers based on side
To filtering thin line feature recognizer and based on Hessian matrix thin line feature recognizer ROC curve comparison diagrams；
Figure 13 is that vein curvature gray feature and background curvature gray feature fusion recognition algorithm are refused sincere with threshold value to melt
Close the change curve of weight α；
Figure 14 is to merge the ROC curve of background curvature gray feature recognizer based on vein curvature gray feature and be based on
The ROC curve comparison diagram of independent vein curvature gray feature recognizer；
Figure 15 is that three kinds of characteristic threshold value fusion recognition algorithms refuse the sincere change curve with Threshold Fusion weight beta；
Figure 16 is the recognizer ROC curve that three kinds of characteristic threshold values merge and is identified based on independent vein curvature gray feature
The ROC curve comparison diagram of algorithm；
Specific embodiment
Specific embodiments of the present invention are further described below in conjunction with the accompanying drawings.
The finger vein identification method based on multiple features Threshold Fusion of the present embodiment, includes the following steps：
σ takes 1.5, t to take 4 in S1, formula (1), and the template window length of side is p=13, is configured to the square of one 13 × 13
Dimensional Gaussian template.
S2, two-dimensional Gaussian function x directions first derivative h ' is obtained respectivelyxThe first derivative h ' in (x, y), y directionsy(x,y)、x
Direction second-order partial differential coefficient h "xx(x, y), y directions second-order partial differential coefficient h "yy(x, y) and second-order mixed partial derivative h "xy(x,y)。
S3, each derivative expressions that template window coordinate is substituted into step S2 are obtained into 5 partial derivatives of template.
S4,5 partial derivative convolution finger vena original images with template, original image is as shown in Figure 1, convolution process
In, original image matrix expansion is expanded using boundary value replica method.The result that convolution obtains is 5 local derviations of image array
Number.
S5,5 partial derivatives of the obtained image arrays of step S4 are substituted into directional derivative formula (2) (3), obtains image moment
4 directional derivatives of battle array.
S6, it 4 directional derivatives that step S5 is obtained is substituted into curvature estimation formula i.e. formula (4) one by one obtains image array and exist
Curvature feature figure on four direction, as shown in Figure 2-5.
Curvature on S7, the four direction for obtaining step S6 substitutes into formula (5), acquires on the four direction of image array most
Deep camber figure, as shown in Figure 6.The curvature gray level image of Fig. 6 be dimensional Gaussian template variance be 1.5 when extraction as a result, by Fig. 6
As it can be seen that the curvature gray level image medium sized vein details of variance hour extraction is more, but thick vein profile can not complete extraction go out
Come.
σ takes 2.5, variance of unit weight t to take 4 in S8, formula (1), and the template window length of side is p=21, is configured to one 21 × 21
Square dimensional Gaussian template, and repeat step S2-S7, obtain variance as shown in Figure 7 be 2.5 curvature gray level image carry
Take result.As seen from Figure 7, compared with Fig. 6, thick vein profile information is more complete in the extraction result of big variance, but details
Veinprint becomes unintelligible.
S9, the curvature gray feature figure that the standard deviation for obtaining S8 and S7 is 1.5 and 2.5 are merged according to formula (6), are melted
The results are shown in Figure 8 for conjunction.The big region of the curvature gray scale extracted in Fig. 8 corresponds to the venosomes of finger venous image, can
See that the result for having merged two standard deviations compared with single standard deviation extraction result, that is, contains the integrity profile of thick vein,
More details veinprint is contained again, achieves preferable syncretizing effect.
S10, the curvature gray feature of fusion according to formula (7) (8) is decomposed, venosomes curvature is obtained by formula (8)
Gray feature is as shown in figure 9, to obtain background area curvature feature by formula (7) as shown in Figure 10.White area is in Fig. 9 and Figure 10
The region larger for curvature value, by the normalized of formula (7), the curvature value range of background area curvature gray feature is also
0-255, normalized are the threshold calculations and Threshold Fusion of subsequent step for convenience.
S11, the curvature gray feature of fusion is divided into binaryzation by threshold, connected region filtering is tabled look-up
It is as shown in figure 11 that refinement obtains curvature thin line feature.As seen from Figure 11, curvature thin line feature is one finger venous image of reflection
The simplest feature of vein topological structure contains only the single pixel trunk information of venosomes.
S12, threshold calculations and matching verification.Two width finger venous images have been performed both by step S1-S11 and have obtained two width hands
Refer to three pairs of features, that is, vein curvature gray feature, background curvature gray feature and curvature thin line feature of vein image.Vein is bent
Rate gray feature and background curvature gray feature carry out threshold calculations according to formula (9) and obtain threshold value T1And T0, curvature thin line feature presses
Illuminated (10) (11) carries out threshold calculations and obtains threshold value T2, fusion threshold value T is obtained according still further to Threshold Fusion formula, that is, formula (12), and
Current comparison threshold value and the recognition threshold T under some accuracy of system identification calculated by image library by comparingsTo judge two width fingers
Whether vein image comes from same root finger.
It is the analysis of experimental data for carrying algorithm to the present invention based on image data base, threshold value weight α and β below.
And the experimental result for being compared the performance of performance and other finger vena recognizers of the invention for carrying algorithm.
This algorithm simulating image is that 500 classes of the venous collection equipment acquisition of this laboratory research and development refer to vein image, often
Class image has 3 width figures, and image original size is 500 × 220, and in order to save image processing time, image first is used bilinearity
Interpolation is normalized to 160 × 64.Personnel's normal use equipment is collected during acquisition, a finger using three times, protect automatically by equipment
Deposit three width images.Gatherer process time span is long, and acquisition environmental change is complicated, and collected personnel amount is more, meets current big portion
The practical application scene divided, therefore image library conclusion tool has significant practical applications.Algorithm routine is in MATLAB R2014a
On write and run, Simulated computer memory 4G, basic frequency 3.4GHZ, operating system are 64 Windows7.Accuracy of system identification (False
Acceptance Rate, FAR) it is the finger vein image that different fingers acquire respectively, carrying out 1:It is judged as during 1 matching same
Ratio shared by one finger；It is the finger vein figure that same finger acquires respectively to refuse sincere (False Rejection Rate, FRR)
Picture is carrying out 1:The ratio shared by different fingers is judged as during 1 matching.This emulation experiment 1:The 1 different finger classes calculated
Between images match number be 1122750 times, images match number is 1500 times in same finger class.
By comparing FAR-FRR curves, that is, ROC (receiver operator characteristic of different characteristic
Curve, ROC) curve is identified the comparison of performance, and for ROC curve close to reference axis, performance is better.The present invention is to being proposed
The hand vein recognition algorithm based on multiple features Threshold Fusion emulate, while in order to compare, to the filament based on trend pass filtering
Feature recognition algorithms and thin line feature recognizer based on Hessian matrixes emulate.
Figure 12 is to extract three kinds of features herein algorithm and other recognizer ROC curves is individually identified.It can by Figure 12
Know, it is proposed by the present invention to be based on side based on vein curvature gray feature recognizer and based on curvature thin line feature recognizer ratio
There is better performance to filtering thin line feature recognizer and based on Hessian matrix thin line features recognizer.This be by
In having better humidification to image with dimensional Gaussian formwork calculation curvature.Due to vein curvature gray feature include it is useful
Information is more than single pixel curvature thin line feature, so vein curvature gray feature performance is slightly above curvature thin line feature.Due to the back of the body
Scape curvature gray feature contains only a small amount of vein image available information, so background curvature gray feature performance is less than vein
Curvature gray feature and curvature thin line feature, help out in fusion recognition algorithm.
According to formula (12), the Threshold Fusion of two region curvature gray features is first carried out, it in FAR is 10 that Figure 13, which is,-6Shi Liang areas
Curve is identified after the curvature gray feature Threshold Fusion of domain.It is maximum to promote amplitude by α FRR in 20%-30% as shown in Figure 13.With reference to
Behavior pattern during other FAR, weight α are chosen for 23%, are 10 in FAR-6When FRR be reduced to 19.73% from 24.19%, can
See that fusion recognition performance improves, sincere be substantially reduced is refused after two region curvature gray feature Threshold Fusions under low accuracy of system identification.
Figure 14 is the ROC curve of two region curvature gray feature Threshold Fusion recognizers, as seen from Figure 14, bent by 23% background
After the weighting of rate gray feature threshold value, the ROC curve of two region curvature gray feature fusion recognition algorithms is closer to reference axis, explanation
Recognition performance improves after fusion.New threshold value after fusion is known as two curvature gray features fusion threshold value.
Two curvature gray features fusion threshold value with curvature thin line feature threshold value is merged again, determines weight beta.Figure 15 is
FAR is 10-6When final fusion FRR with the change curve of weight beta, as seen from Figure 15, β is minimum in 40% or so FRR, with reference to it
Behavior pattern during his FAR selects β as 40%.It is 10 in FAR-6When the fusion of three kinds of characteristic threshold values FRR reduced from 19.73%
To 8.13%.Figure 16 is that the ROC curve of three kinds of characteristic threshold value fusion recognition algorithms and vein curvature gray feature are individually identified
The ROC curve comparison of algorithm.Recognition performance is promoted significantly after Threshold Fusion as shown in Figure 16, is 10 particularly in accuracy of system identification-6When,
Sincere 24.19% promotion from single vein curvature gray feature recognizer is refused to based on multiple features Threshold Fusion recognizer
8.13%, illustrate that recognition performance is promoted apparent under low accuracy of system identification.Final fusion threshold value is T=T0×0.138+T1×0.462+
T2× 0.4, each Threshold Fusion weight meets theory analysis.This is because vein curvature gray feature and curvature thin line feature include
The main information of vein image, plays main function, so threshold value T in Threshold Fusion recognizer1,T2Institute's accounting in T
Example up to 46.2% and 40%；And background curvature gray feature generally only has a small amount of information of vein image, only in vein area
Main recognition reaction can be just played in the low-quality image that domain obscures, and in the fusion recognition based on whole image library calculated performance
It only plays auxiliary to use, so threshold value T0Proportion only has 13.8% in T.Image library is to simulate used by present invention experiment
Collecting device practical service environment collects, therefore the weight is to the finger venous collection equipment researched and developed based on this laboratory
Finger vein recognition system has applicability.
The embodiment of the present invention is explained in detail above in conjunction with attached drawing, but the present invention is not limited to above-mentioned implementations
Example, within the knowledge of a person skilled in the art, can also make under the premise of present inventive concept is not departed from
Go out to obtain various change, also should be regarded as protection scope of the present invention.
Claims (3)
1. a kind of finger vein identification method based on multiple features Threshold Fusion, it is characterised in that include the following steps：
The dimensional Gaussian template of step 1, the multiple standard deviations of construction, Gaussian template expression formula are as follows：
In formula (1), x and y values are codetermined by two-dimensional Gaussian function standard deviation sigma and Gaussian template Plays difference factor t, two dimension
The Gaussian template square window length of side is p=2 × t × σ+1, therefore x and y values are [- p, p].σ values are variable, and certainly
Determine the size of window, different windows size corresponds to the vein of extraction different thicknesses；
Step 2, the first-order partial derivative h' that dimensional Gaussian template is first obtainedx(x,y)h'y(x, y), second-order partial differential coefficient h "xx(x,y)
h″yy(x, y), mixed partial derivative h "xy(x, y), and convolved image matrix obtains image array respectively with this 5 partial derivative templates
5 partial derivative H'x,H'y,H'xx,H'yy,H′xy, then ask with directional derivative formula single order, the Second order directional of four angles；
Reference axis x-axis direction is 0 °, and y-axis is 90 °；
In formula (2) (3)Take 0 °, 45 °, 90 °, 135 °.Directional derivative is substituted into curvature formulations (4), obtains vein image in each side
Upward corresponding curvature value, and four direction maximum curvature value is asked to save as the final curvature value of vein image, such as formula (5):
K is the curvature feature finally extracted in formula (5), is maximum curvature value on four direction；
Step 3 changes standard deviation sigma, and the extraction result for choosing suitable multiple standard deviations carries out curvature Fusion Features；Fusion formula
As shown in formula (6)：
K in formula (6)σ1σ is taken for standard deviation1Dimensional Gaussian template extraction curvature as a result, kσ2σ is taken for standard deviation2Two dimension it is high
The curvature of this template extraction is as a result, K is the fusion results that two standard deviations extract result；
Step 4, curvature gray feature decompose, and normalize curvature feature respectively according to positive negative region according to formula (7) (8) and are carried on the back
Scape curvature gray feature K0 and vein curvature gray feature K1；
In formula (7) (8), K is curvature gray level image, and K_min and K_max are according to from small respectively by curvature gray level image matrix value
To two vectors that are big and sorting from big to small；The average value of preceding 10 values of K_min is taken as the normalized upper limit of negative region,
Taking the average value of preceding 10 values of K_max, the exposure for mitigating image border is mutated to curvature as the normalized upper limit of positive region
As a result influence；
Step 5, the extraction of curvature thin line feature, curvature gray level image K pass through binaryzation, filtering, and refinement obtains thin line feature；
Threshold calculations are individually identified in step 6, curvature gray feature
Background curvature gray feature and vein curvature gray feature belong to curvature gray feature, the phase relation of this paper calculating matrix
Count the similarity to represent curvature gray feature；Related coefficient calculation formula such as formula (9):
K in formula (9)a,KbIt is meant for two by curvature gray feature square Cov and calculates two curvature gray feature Ka,KbCovariance,
Var, which is meant, calculates two curvature gray feature Ka,KbVariance；It is T to remember background curvature gray feature recognition threshold0, vein curvature
Gray feature recognition threshold is T1；
Threshold calculations are individually identified in step 7, curvature thin line feature
Curvature thin line feature threshold value is calculated using modified Hausdorff distance algorithms；For A and two curvature thin line features of B,
MHD calculation formula such as formula (10) (11):
T2(A, B)=max (d (A, B), d (B, A)) (11)
D is the minimum average B configuration distance of A to B in formula (10), T in formula (11)2MHD threshold values for A and B；First calculate certain point A in AmTo B
Middle all the points BnDistance, wherein 1≤n≤N is total to N number of point；Then minimum value is obtained, traverse all M points in A and is averaging
Value；It is come again from B to A again, takes the two higher value as the MHD threshold values between A and B；
Step 8, three kinds of characteristic threshold value T0,T1And T2Fusion obtains the final threshold value T after fusion；
Step 9, match cognization：
Threshold value is bigger to represent that image similarity is higher, identification verification expression formula such as formula (13):
T in formula (13)sFor the recognition threshold being calculated according to image library, different accuracy of system identification is corresponding with different recognition thresholds,
T is the matched final threshold value that two width finger venous images calculate, and can judge two width finger venous images according to formula (13)
Whether same finger is derived from.
2. a kind of finger vein identification method based on multiple features Threshold Fusion according to claim 1, it is characterised in that
The realization of fusion described in step 8, it is specific as follows：
Since curvature gray feature and curvature thin line feature are the features of different dimensions, threshold calculations mode is different, therefore three
Kind Threshold Fusion sequence is the Threshold Fusion for first carrying out two region curvature gray features, obtains the best weight α of the two and new threshold
Value, then obtain weight beta and final threshold value T, expression formula such as formula (12) with curvature thin line feature Threshold Fusion:
T=(1- β) (α T0+(1-α)T1)+βT2 (12)
T in formula (12)0,T1,T2Respectively background curvature gray feature, vein curvature gray feature and curvature thin line feature are returned
One changes hundred-mark system threshold value, and α is threshold value T0,T1Fusion weight, β be threshold value T2With T0,T1Merge the fusion weight of threshold value.
3. a kind of finger vein identification method based on multiple features Threshold Fusion according to claim 1, it is characterised in that
Step curvature thin line feature is extracted, and is implemented as follows：
Binary conversion treatment uses threshold partitioning algorithm, threshold value is obtained using the mean value in window and variance, to curvature
Gray-scale map K carries out binaryzation, and curvature thin line feature is obtained using connected region filtering and noise reduction, then with refinement method of tabling look-up.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810015658.8A CN108256456B (en) | 2018-01-08 | 2018-01-08 | Finger vein identification method based on multi-feature threshold fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810015658.8A CN108256456B (en) | 2018-01-08 | 2018-01-08 | Finger vein identification method based on multi-feature threshold fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108256456A true CN108256456A (en) | 2018-07-06 |
| CN108256456B CN108256456B (en) | 2020-04-07 |
Family
ID=62725975
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810015658.8A Active CN108256456B (en) | 2018-01-08 | 2018-01-08 | Finger vein identification method based on multi-feature threshold fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108256456B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109271966A (en) * | 2018-10-15 | 2019-01-25 | 广州广电运通金融电子股份有限公司 | A kind of identity identifying method, device and equipment based on finger vein |
| CN109902586A (en) * | 2019-01-29 | 2019-06-18 | 平安科技（深圳）有限公司 | Palmmprint extracting method, device and storage medium, server |
| CN110008902A (en) * | 2019-04-04 | 2019-07-12 | 山东财经大学 | A kind of finger vein identification method and system merging essential characteristic and deformation characteristics |
| CN110147769A (en) * | 2019-05-22 | 2019-08-20 | 成都艾希维智能科技有限公司 | A kind of finger venous image matching process |
| CN110163123A (en) * | 2019-04-30 | 2019-08-23 | 杭州电子科技大学 | One kind referring to vein fusion identification method based on single width near-infrared finger-image fingerprint |
| CN110188614A (en) * | 2019-04-30 | 2019-08-30 | 杭州电子科技大学 | It is a kind of based on skin crack segmentation NLM filtering refer to vein denoising method |
| CN110502996A (en) * | 2019-07-22 | 2019-11-26 | 杭州电子科技大学 | A kind of dynamic identifying method towards fuzzy finger vein image |
| CN110569815A (en) * | 2019-09-12 | 2019-12-13 | 南京邮电大学 | Finger vein image feature extraction algorithm based on improved gray level non-uniform correction |
| CN110751029A (en) * | 2019-09-12 | 2020-02-04 | 南京邮电大学 | Maximum curvature-based adaptive finger vein line extraction method |
| CN110880003A (en) * | 2019-10-12 | 2020-03-13 | 中国第一汽车股份有限公司 | Image matching method and device, storage medium and automobile |
| CN111612083A (en) * | 2020-05-26 | 2020-09-01 | 济南博观智能科技有限公司 | Finger vein identification method, device and equipment |
| CN111639555A (en) * | 2020-05-15 | 2020-09-08 | 圣点世纪科技股份有限公司 | Finger vein image noise accurate extraction and self-adaptive filtering denoising method and device |
| CN111950405A (en) * | 2020-07-30 | 2020-11-17 | 内蒙古智诚物联股份有限公司 | Vein recognition input and output device based on artificial intelligence and recognition method thereof |
| CN114023249A (en) * | 2021-10-29 | 2022-02-08 | 卡莱特云科技股份有限公司 | LED display screen image light point extraction method and device and LED display screen correction method |
| CN115272684A (en) * | 2022-09-29 | 2022-11-01 | 山东圣点世纪科技有限公司 | Method for processing pseudo noise in vein image enhancement process |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2011052085A1 (en) * | 2009-10-30 | 2011-05-05 | 富士通フロンテック株式会社 | Biometric information registration method, biometric authentication method, and biometric authentication device |
| CN104239769A (en) * | 2014-09-18 | 2014-12-24 | 北京智慧眼科技发展有限公司 | Identity recognition method and system based on finger vein characteristics |
| CN105184272A (en) * | 2015-09-21 | 2015-12-23 | 中国人民解放军国防科学技术大学 | Hand vein recognition method based on curve matching |
| CN106611168A (en) * | 2016-12-29 | 2017-05-03 | 杭州电子科技大学 | Fast finger vein recognition method based on thinned images and direction field patterns |
-
2018
- 2018-01-08 CN CN201810015658.8A patent/CN108256456B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2011052085A1 (en) * | 2009-10-30 | 2011-05-05 | 富士通フロンテック株式会社 | Biometric information registration method, biometric authentication method, and biometric authentication device |
| CN104239769A (en) * | 2014-09-18 | 2014-12-24 | 北京智慧眼科技发展有限公司 | Identity recognition method and system based on finger vein characteristics |
| CN105184272A (en) * | 2015-09-21 | 2015-12-23 | 中国人民解放军国防科学技术大学 | Hand vein recognition method based on curve matching |
| CN106611168A (en) * | 2016-12-29 | 2017-05-03 | 杭州电子科技大学 | Fast finger vein recognition method based on thinned images and direction field patterns |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109271966A (en) * | 2018-10-15 | 2019-01-25 | 广州广电运通金融电子股份有限公司 | A kind of identity identifying method, device and equipment based on finger vein |
| CN109271966B (en) * | 2018-10-15 | 2021-10-26 | 广州广电运通金融电子股份有限公司 | Identity authentication method, device and equipment based on finger veins |
| CN109902586A (en) * | 2019-01-29 | 2019-06-18 | 平安科技（深圳）有限公司 | Palmmprint extracting method, device and storage medium, server |
| CN110008902B (en) * | 2019-04-04 | 2020-11-17 | 山东财经大学 | Finger vein recognition method and system fusing basic features and deformation features |
| CN110008902A (en) * | 2019-04-04 | 2019-07-12 | 山东财经大学 | A kind of finger vein identification method and system merging essential characteristic and deformation characteristics |
| CN110163123A (en) * | 2019-04-30 | 2019-08-23 | 杭州电子科技大学 | One kind referring to vein fusion identification method based on single width near-infrared finger-image fingerprint |
| CN110188614A (en) * | 2019-04-30 | 2019-08-30 | 杭州电子科技大学 | It is a kind of based on skin crack segmentation NLM filtering refer to vein denoising method |
| CN110188614B (en) * | 2019-04-30 | 2021-03-30 | 杭州电子科技大学 | NLM filtering finger vein denoising method based on skin crack segmentation |
| CN110163123B (en) * | 2019-04-30 | 2021-02-26 | 杭州电子科技大学 | Fingerprint finger vein fusion identification method based on single near-infrared finger image |
| CN110147769A (en) * | 2019-05-22 | 2019-08-20 | 成都艾希维智能科技有限公司 | A kind of finger venous image matching process |
| CN110147769B (en) * | 2019-05-22 | 2023-11-07 | 成都艾希维智能科技有限公司 | Finger vein image matching method |
| CN110502996B (en) * | 2019-07-22 | 2022-03-15 | 杭州电子科技大学 | Dynamic identification method for fuzzy finger vein image |
| CN110502996A (en) * | 2019-07-22 | 2019-11-26 | 杭州电子科技大学 | A kind of dynamic identifying method towards fuzzy finger vein image |
| CN110751029A (en) * | 2019-09-12 | 2020-02-04 | 南京邮电大学 | Maximum curvature-based adaptive finger vein line extraction method |
| CN110569815A (en) * | 2019-09-12 | 2019-12-13 | 南京邮电大学 | Finger vein image feature extraction algorithm based on improved gray level non-uniform correction |
| CN110751029B (en) * | 2019-09-12 | 2022-08-02 | 南京邮电大学 | Maximum curvature-based adaptive finger vein line extraction method |
| CN110880003A (en) * | 2019-10-12 | 2020-03-13 | 中国第一汽车股份有限公司 | Image matching method and device, storage medium and automobile |
| CN110880003B (en) * | 2019-10-12 | 2023-01-17 | 中国第一汽车股份有限公司 | Image matching method and device, storage medium and automobile |
| CN111639555A (en) * | 2020-05-15 | 2020-09-08 | 圣点世纪科技股份有限公司 | Finger vein image noise accurate extraction and self-adaptive filtering denoising method and device |
| CN111612083A (en) * | 2020-05-26 | 2020-09-01 | 济南博观智能科技有限公司 | Finger vein identification method, device and equipment |
| CN111950405A (en) * | 2020-07-30 | 2020-11-17 | 内蒙古智诚物联股份有限公司 | Vein recognition input and output device based on artificial intelligence and recognition method thereof |
| CN114023249A (en) * | 2021-10-29 | 2022-02-08 | 卡莱特云科技股份有限公司 | LED display screen image light point extraction method and device and LED display screen correction method |
| CN115272684A (en) * | 2022-09-29 | 2022-11-01 | 山东圣点世纪科技有限公司 | Method for processing pseudo noise in vein image enhancement process |
| CN115272684B (en) * | 2022-09-29 | 2022-12-27 | 山东圣点世纪科技有限公司 | Method for processing pseudo noise in vein image enhancement process |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108256456B (en) | 2020-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108256456A (en) | A kind of finger vein identification method based on multiple features Threshold Fusion | |
| Jain et al. | Intelligent biometric techniques in fingerprint and face recognition | |
| Jiang et al. | Detecting the fingerprint minutiae by adaptive tracing the gray-level ridge | |
| US7599530B2 (en) | Methods for matching ridge orientation characteristic maps and associated finger biometric sensor | |
| EP1339008B1 (en) | Authentication method, and program and apparatus therefor | |
| CN1152340C (en) | Fingerprint image enhancement method based on knowledge | |
| Hambalık | Fingerprint recognition system using artificial neural network as feature extractor: design and performance evaluation | |
| CN104239769A (en) | Identity recognition method and system based on finger vein characteristics | |
| CN105975905B (en) | A kind of finger vena method for quickly identifying | |
| CN111914616A (en) | Finger vein identification and anti-counterfeiting integrated method, device, storage medium and equipment | |
| CN111223063A (en) | Finger vein image NLM denoising method based on texture features and binuclear function | |
| Al-Nima et al. | Segmenting finger inner surface for the purpose of human recognition | |
| CN114973308A (en) | Finger vein identification method and system based on elastic weight solidification and multivariate similarity loss | |
| Ghoualmi et al. | A multi-objective optimization approach for palm vein feature selection based on the discrete PSO | |
| Fang et al. | Deep belief network based finger vein recognition using histograms of uniform local binary patterns of curvature gray images | |
| Rezaei et al. | Persian signature verification using fully convolutional networks | |
| Prashanth et al. | DWT based offline signature verification using angular features | |
| Sarfraz | Introductory chapter: On fingerprint recognition | |
| Szymkowski et al. | A novel approach to fingerprint identification using method of sectorization | |
| CN115439891A (en) | Fingerprint identification method based on small fingerprint head, low computational power and low memory chip | |
| CN115204238A (en) | PPG signal identity recognition method for wearable equipment and wearable equipment | |
| CN114973307A (en) | Finger vein identification method and system for generating countermeasure and cosine ternary loss function | |
| Nasiri et al. | A fuzzy approach for the automatic off-line signature verification problem base on geometric features | |
| Ramachandra et al. | Signature verification using graph matching and cross-validation principle | |
| Ramos et al. | Analysis and improvements of fingerprint enhancement from gabor iterative filtering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |